Predicting Neoadjuvant Chemotherapy Response in Triple-Negative Breast Cancer Using Pre-Treatment Histopathologic
Hikmat Khan1, Ziyu Su1, Huina Zhang2
1Department of Pathology, College of Medicine, Wexner Medical Center, The Ohio State University, Columbus, OH, 43210, USA.
Arxiv
|July 31, 2025
Summary
An AI model predicts triple-negative breast cancer chemotherapy response using H&E slides. This approach enhances personalized treatment and identifies key immune biomarkers for better patient outcomes.
Area of Science:
- Computational pathology
- Digital pathology
- Precision oncology
Background:
- Triple-negative breast cancer (TNBC) presents significant clinical challenges due to its aggressive nature and limited targeted therapies.
- Predicting neoadjuvant chemotherapy (NACT) response is crucial for tailoring treatment and improving outcomes in TNBC patients.
Purpose of the Study:
- To develop and validate an attention-based multiple instance learning (MIL) framework for predicting pathologic complete response (pCR) in TNBC.
- To enhance the interpretability of the predictive model by correlating attention maps with immune cell infiltration.
Main Methods:
- An attention-based MIL framework was developed to analyze pre-treatment H&E-stained biopsy slides.
- The model was trained on a cohort of 174 TNBC patients and validated on an independent cohort of 30 patients.
- Attention maps were spatially co-registered with multiplex immunohistochemistry (mIHC) data for PD-L1, CD8+ T cells, and CD163+ macrophages.
Main Results:
- The MIL model achieved a mean AUC of 0.85 (cross-validation) and 0.78 (external testing), demonstrating robust predictive performance.
- Attention regions showed moderate spatial overlap with immune-enriched areas (IoU: 0.47 for PD-L1, 0.45 for CD8+, 0.46 for CD163+).
- This overlap suggests biological relevance of immune biomarkers in predicting NACT response.
Conclusions:
- The developed MIL framework accurately predicts NACT response in TNBC from H&E slides.
- Model interpretability analysis links predictive regions to immune biomarkers, supporting their role in treatment response.
- This approach advances precision oncology by enabling accurate response prediction and identifying potential histological biomarkers.
Keywords:
artificial intelligence (AI)neoadjuvant chemotherapy (NACT)pathologic complete response (pCR)treatment response predictiontriple-negative breast cancer (TNBC)
